Many models are too complex for walkthroughs to be interactive, which requires an update rate of at least 20 frames/second. If, for example, one has a graphics engine capable of rendering 2 million polygons/second, models of more than 100 thousand polygons cannot be smoothly traversed by brute force alone. Over the 1995-96 year, we developed a variety of techniques for making interactive walkthroughs of massive models possible. During 1996-97, we studied how to integrate several of these techniques into a single rendering framework. This is non-trivial, because in general each technique needs its own particular data structure; these are incompatible; and machine memories cannot contain multiple data structures of a massive model. We achieved considerable success. In August, at the annual SIGGRAPH meeting, we demonstrated interactive walkthrough, at speeds of 4-20 frames/second, of a massive model of some 13 million polygons. One of the techniques used was to fill the model space with some 10,000 cells that lay along the paths most viewers would want to take. For each cell, a box, some 6x cell size on each side, was constructed. All the geometry outside the box is rendered onto meshes associated with each box wall, using projective textures, as a pre-process. At run time, only the simplified projective textures are rendered for all the geometry outside the box the viewer currently inhabits. The geometry inside the box is rendered using level-of-detail simplfication as described in Subproject 9 above. Only the geometry in the view frustum is rendered; a hierarchical culling technique enables these decisions to be made very rapidly. Yet another hierarchical culling technique is used to avoid rendering objects that are occluded by single objects or groups of objects nearer the viewer's eye.